Healthcare Resources Allocation: Targeting "The Right Intervention to the Right Population"

Description

This proposal addresses a class of Precision Public Health problems from an operations research perspective. Simply put, Precision Public Health means "implementing the right intervention to the right population." As healthcare resources are scarce, allocating them the most effectively is essential. On the one hand, even in the era of big data, it remains challenging to identify the "right population" to be targeted for an intervention/treatment that has been clinically proven to be effective. This is because, very often, a (clinical or non-clinical) research study typically only reports the evidence at high levels, such as the demographic features of its study population and estimates of the average benefit. On the other hand, knowledge of the heterogeneous effect is essential to the above notation of "Precision Public Health" and the resource allocation problem. For example, patients of chronic obstructive pulmonary disease (COPD) can benefit from taking certain vaccines. The average benefit of a vaccine would be measured in terms of reduced hospitalization in the ensuing months. Benefits from different vaccines were reported by demographic profiling, such as sex, age groups (e.g., over/below 65 years old), severity, and comorbidity. However, it is unknown how an individual could benefit from a vaccine. Take a male COPD patient outside of a study sample as an example, who is 70 years old and of mild severity but with comorbidity. The study might report several average benefits, including that for his age group - 65+ years, the average for males of all ages, another average for those with comorbidity conditions, and so on. Nevertheless, the range of the benefits for him remains unknown. Even with the "raw data" available so that all individuals' information in the study is accessible, the points of observation usually do not entail any statistically significant inference due to small sample sizes by virtue of expensive clinical or nonclinical trials. To overcome this challenge, we will develop a novel approach, which is built upon the intuition that “similar” patients in "features" (profiling) respond to an intervention similarly. Specifically, we propose an evidence-based efficacy approximation that incorporates patients' profiles by a kernel-weighting method to capture the patient-specific efficacy. The proposed weighting method is then embedded into a robust optimization model, from which intervention recommendations are made to the most suitable candidates to maximize the effectiveness of the aggregate intervention.